“Network effects” can describe several different mechanisms. A messaging product, a marketplace and a shared fraud-detection service do not become more useful in the same way.
Classification is useful when it tells an investor what to test. The framework below separates the claimed mechanism from the quality of the evidence. A plausible type can still have weak support; a modest, well-demonstrated effect can be more valuable than an elaborate untested story.
Classify the mechanism before scoring it
| Working category | Claimed source of additional value | First question to investigate |
|---|---|---|
| Direct interaction | More relevant users can interact with existing users | Are the additional participants people existing users need? |
| Cross-side interaction | More of one participant group improve another group's experience | Does the additional supply or demand produce better matches? |
| Platform and complements | More useful complementary products attract users and support developers | Are the complements used, maintained and difficult to replace? |
| Shared data improvement | Participation contributes information that improves outcomes for others | Does additional usable data measurably improve the product? |
These are working categories, not mutually exclusive boxes. A software platform can combine direct collaboration, third-party tools and a shared-data service. Assess each claim separately so that one demonstrated effect does not validate all the others.
Direct effects: relevance matters more than membership
Imagine a secure messaging product used by emergency-response teams. Adding relevant colleagues could improve coordination. Adding thousands of unrelated users would not necessarily help the existing team.
The diligence record should therefore identify the relevant group, the activity improved and the limit of the effect. Ask whether a single team gets enough value to adopt independently, or whether adoption requires several organisations to move together.
A useful conclusion might be “direct effect within an active team; little evidence of value across unrelated teams”. That boundary matters when estimating expansion into new customers.
Cross-side effects: assess both directions
For a specialist equipment marketplace, more suitable listings may attract buyers, while more qualified buyers may attract sellers. The directions can differ in strength.
Rochet and Tirole's platform analysis highlights the need to attract both sides and consider how prices are allocated between them. Its economic model is a starting point for reasoning, not empirical proof about a particular marketplace. Rochet and Tirole.
In the worksheet, record the effect on each side separately. More competing sellers might improve buyer choice while reducing each seller's chance of a transaction. A platform must manage that tension rather than treat every additional participant as equally beneficial.
Platform effects: count useful complements
A large integration directory can exaggerate the value of an ecosystem. Some integrations may be unused, redundant or no longer maintained.
Ask which complements influence purchasing or retention, who maintains them and whether developers earn enough to keep contributing. Then examine dependence: could customers obtain the same tools through a competing platform with little effort?
The evidence should connect complement availability to a user outcome. A list of partner logos establishes relationships, not the magnitude of the effect.
Data effects: establish the complete improvement loop
For a shared fraud-detection service, the proposed loop might be: participating customers contribute lawful, usable observations; the service learns from them; detection improves for other customers; improved performance attracts further participation.
Each step needs support. Additional data can be repetitive, poorly labelled or unsuitable for reuse. Product improvement might come from engineering work rather than from network growth.
Request comparisons that isolate the contribution of additional data as far as practical. Check permissions, error rates and whether the improvement persists on new cases. Treat this as a data-improvement hypothesis until the link is demonstrated.
Grade confidence and record the limit
Use a simple evidence scale consistently:
| Grade | Evidence available |
|---|---|
| Unproven | A plausible explanation without supporting outcome data |
| Indicative | Observed relationships with important alternative explanations remaining |
| Supported | Repeated comparable evidence, with major confounders examined |
| Strongly tested | Credible experiments or other rigorous tests supporting the stated mechanism in its defined setting |
These grades describe confidence, not the commercial size of the benefit. Add separate fields for magnitude, geographic or customer boundary, ongoing subsidy and the next test.
A completed entry might read: “Cross-side effect, indicative; faster fulfilment in dense categories; subsidy and customer-tenure differences unresolved.” Our guide to measuring network strength develops those tests. The framework's value is a precise investment question, not a badge that declares the company defensible.
FAQ
What are the four types of network effects in this framework?
The working categories are direct interaction, cross-side interaction, platforms and complements, and shared data improvement. They organise different claims about how participation creates value. They are not an exhaustive or mutually exclusive scientific taxonomy; one business may combine several mechanisms.
What makes a network effect strong?
Assess the size of the benefit, the relevant network boundary, durability and the evidence supporting the mechanism. Growth in unrelated participants may contribute little. A strong finding connects additional participation to better outcomes after considering subsidies, concentration and competing explanations.
Are platform integrations automatically a network effect?
No. Determine whether useful complements attract or retain users and whether that participation supports further complement creation. Unused integrations and partner logos do not establish the loop. Look for adoption, maintenance and outcomes that would be harder to obtain through a competing platform.
Can a network effect reverse?
Yes. Departures can reduce value for remaining participants and encourage further departures, while congestion or low-quality additions can also weaken the experience. The speed and extent depend on the mechanism. Continue monitoring engagement and quality after an initial effect has been demonstrated.
